AMD CEO Teases Zen Processor Family, Summit Ridge Makes An Appearance

[somebodyspecial]

ZEN die size had better be MUCH larger than Intel's CPU side currently, or it will get priced to death no matter what by Intel. You need a KING, just like Dirk said ~2011. I really hope AMD looked at 300mm^2 and shot for that. If it's less than 200 this whole thing is pointless and will just be a better crappy chip. Profits come from pricing power, not good enough - but cheaper than Intel. With Intel concentrating so hard on the gpu side (basically ignoring enthusiasts) for so long, this is AMD's last chance to put out a KING and price it to make a billion again.

Again, I really hope management just looked at engineering and said, "look make me the fastest QUAD you can for 300mm^2 so we can price to actually make some cash for a change! and if you come back with something under 250 you're all FIRED!". That would make some profits However, if management told them to make something small or even close to Intel cpu side, they should all resign 😉 The point of this design should have been WINNING at all costs, not close but cheaper which is easily priced to bargain bin for years if you're Intel.

 

[Robert Cook]

ZEN die size had better be MUCH larger than Intel's CPU side currently, or it will get priced to death no matter what by Intel. You need a KING, just like Dirk said ~2011. I really hope AMD looked at 300mm^2 and shot for that. If it's less than 200 this whole thing is pointless and will just be a better crappy chip. Profits come from pricing power, not good enough - but cheaper than Intel. With Intel concentrating so hard on the gpu side (basically ignoring enthusiasts) for so long, this is AMD's last chance to put out a KING and price it to make a billion again.

Again, I really hope management just looked at engineering and said, "look make me the fastest QUAD you can for 300mm^2 so we can price to actually make some cash for a change! and if you come back with something under 250 you're all FIRED!". That would make some profits However, if management told them to make something small or even close to Intel cpu side, they should all resign 😉 The point of this design should have been WINNING at all costs, not close but cheaper which is easily priced to bargain bin for years if you're Intel.

but is a larger die not less affordable to make? that is why those things tend to cost so much, more surface area means more area to get damaged or come out bad.

 

[TJ Hooker]

but is a larger die not less affordable to make? that is why those things tend to cost so much, more surface area means more area to get damaged or come out bad.

Larger die size also means less chips per wafer, raising cost per chip.

 

[InvalidError]

I really hope AMD looked at 300mm^2 and shot for that. If it's less than 200 this whole thing is pointless and will just be a better crappy chip.

Skylake i5/i7 for LGA1151 are only 122mm^2 including the IGP. If AMD wants to be profitable, they have to beat Skylake on bang-per-buck and they have to do so at a lower silicon cost if possible to improve AMD's gross margin. If AMD went for 300mm^2 for mainstream chips, they would end up outright pricing themselves out of the market.

 

[captaincharisma]

just stick to game consoles AMD. tha'ts the only thing that has worked out for you in the CPU world

 

[Robert Cook]

just stick to game consoles AMD. that's the only thing that has worked out for you in the CPU world

the althlon and phenom series beg to differ. AMD used to beat intel, they still can, they just have lots of work to do.

 

[jimmysmitty]

just stick to game consoles AMD. that's the only thing that has worked out for you in the CPU world

the althalon and phenom series beg to differ. AMD used to beat intel, they still can, they just have lots of work to do.

Athlon*, sorry but it was bugging me.

And it was the Athlon 64 and Athlon 64 X2 that beat Intel. Phenom I was just, well it was just not a good CPU. Phenom II was better but still did not beat Intel. AMD hasn't had a truly competitive CPU since Intel launched Core 2.

 

[alextheblue]

Also, from WCCF Tech and another AMD slide (take this one with a grain of salt):"Compared to AMD’s 'Orochi' quad module, eight core die powering the FX 8350, the Zen based desktop Summit Ridge eight core CPU delivers double the performance in Cinebench R15."

That's not impossible in certain workloads. Remember, the quad-module BD-based designs only have 4 FPUs. An 8 core Zen chip should have 8 FPUs and 16 threads.

excavator only went to their FP1 (mobile high efficiency) APUs. If you compare raw compute between an FX 8350 and an Athlon 5350. The tiny athlon comes in at about 28% that of the 8350 despite being literally 1/4 the CPU. (4 cores @ 2.05Ghz.) And it does this sipping away on a measly 37 watts full system including igpu. The kabini lineup uses excavator architecture with jaguar cores, and is the closest thing to the Zen architecture AMD currently has on the shelves. AMD is calling for a 40% improvement in IPC over that.

No, you're completely confused. Excavator is the successor to Steamroller and thus is BD based. SOME of the FP4 chips such as Carrizo, Bristol Ridge, and Stoney Ridge are Excavator based. However Kabini, Temash, Beema, Mullins, Carrizo-L and every other "cat core" chip (whether it has Bobcat, Jaguar, Puma, or Puma+) are NOT EXCAVATOR. Excavator is Bulldozer derived. None of the cat cores have anything to do with Bulldozer, Steamroller, Piledriver, or Excavator.
The only thing that surprised me is that Stoney Ridge is a 2-core (one module) Excavator (BD-based) chip replacing 4-core Puma+ Carrizo-L designs. That means this is probably the end of the line for mainstream cat cores, though they may still find service in embedded, semi-custom chips, etc.

And it was the Athlon 64 and Athlon 64 X2 that beat Intel.

Longtime Tom's readers remember that the first Athlon (32-bit) ALSO beat Intel. They continued to do so for most of the Athlon and Athlon XP days, as well. There was a few hiccups here and there (like when they got stalled until the Dresden breakthrough) but it wasn't just the Athlon 64 days where they were king.
I wish them luck with Zen, I really hope they're competitive even if they're not the absolute leader in the highest end.

 

[jimmysmitty]

Also, from WCCF Tech and another AMD slide (take this one with a grain of salt):"Compared to AMD’s 'Orochi' quad module, eight core die powering the FX 8350, the Zen based desktop Summit Ridge eight core CPU delivers double the performance in Cinebench R15."

That's not impossible in certain workloads. Remember, the quad-module BD-based designs only have 4 FPUs. An 8 core Zen chip should have 8 FPUs and 16 threads.

excavator only went to their FP1 (mobile high efficiency) APUs. If you compare raw compute between an FX 8350 and an Athlon 5350. The tiny athlon comes in at about 28% that of the 8350 despite being literally 1/4 the CPU. (4 cores @ 2.05Ghz.) And it does this sipping away on a measly 37 watts full system including igpu. The kabini lineup uses excavator architecture with jaguar cores, and is the closest thing to the Zen architecture AMD currently has on the shelves. AMD is calling for a 40% improvement in IPC over that.

No, you're completely confused. Excavator is the successor to Steamroller and thus is BD based. SOME of the FP4 chips such as Carrizo, Bristol Ridge, and Stoney Ridge are Excavator based. However Kabini, Temash, Beema, Mullins, Carrizo-L and every other "cat core" chip (whether it has Bobcat, Jaguar, Puma, or Puma+) are NOT EXCAVATOR. Excavator is Bulldozer derived. None of the cat cores have anything to do with Bulldozer, Steamroller, Piledriver, or Excavator.
The only thing that surprised me is that Stoney Ridge is a 2-core (one module) Excavator (BD-based) chip replacing 4-core Puma+ Carrizo-L designs. That means this is probably the end of the line for mainstream cat cores, though they may still find service in embedded, semi-custom chips, etc.

And it was the Athlon 64 and Athlon 64 X2 that beat Intel.

Longtime Tom's readers remember that the first Athlon (32-bit) ALSO beat Intel. They continued to do so for most of the Athlon and Athlon XP days, as well. There was a few hiccups here and there (like when they got stalled until the Dresden breakthrough) but it wasn't just the Athlon 64 days where they were king.
I wish them luck with Zen, I really hope they're competitive even if they're not the absolute leader in the highest end.

Athlon XP was not bad, no. But I consider Athlon 64 more of beating Intel because it took an entirely different uArch to beat it where as Athlon XP traded blows with Pentium III and 4. The Pentium 4/D on the other hand did not trade blows with the Athlon 64 or X2.

 

[turkey3_scratch]

How come we use instructions per clock instead of clocks per instruction (with lower being better)? It makes more sense, since if you were to create an actual IPC numeric value, I'd think it'd be less than 1 since complex instructions take multiple cycles.

 

[alextheblue]

Athlon XP was not bad, no. But I consider Athlon 64 more of beating Intel because it took an entirely different uArch to beat it where as Athlon XP traded blows with Pentium III and 4. The Pentium 4/D on the other hand did not trade blows with the Athlon 64 or X2.

I said that the FIRST (non-XP) Athlon beat them. There's an old THG article still available on TH about it. Then it continued to do so throughout most (not all) of the Athlon and Athlon XP era. There was a few times where they "traded blows" but outside of heavily optimized Intel compiler produced code that crippled performance on the Athlons? Most of that era was dominated (in terms of performance) by the Athlons. There were periods where they fell behind, but they had the crown at the outset of Athlon and kept it the majority of the time until Intel abandoned Netburst.

 

[Valkrys]

I would assume that Zen will be superior in multithreaded since it will have 8 full cores and 8 logical cores (SMT)

In SMT, neither core is more "virtual" than the other, you have individual physical cores with resources to keep tabs on two or more threads each.

I would assume the FX line up IS AMD's top end to date so the 40% would apply to that overall, but who knows until it's dissected and benched.

Investors need to get excited too, your right.

No, the FX lineup never got updated passed Pile driver. Steamroller, and then excavator only got put into their APUs. Each update giving a small improvement (about 5-10%) but split over different areas. Steamroller was primarily focused on efficiency, and excavator brought in improved IPC.

If you compare the raw compute between an FX 8350, and an Athlon 5350. The 5350 is about 28% of the 8350 using half the cores and half the clock rate (effectively 1/4 the cpu with no L3) and only using 25W of tiny power vs the 8350`s 125w TDP. The 5350 uses excavator architecture with jaguar cores, this is literally the closest thing to zen AMD has on the market currently. And they`re calling for a 40% improvement in IPC.

 

[iam2thecrowe]

Hmm, 40% improvement in ICP..... that could mean a lot, is that per "module"? Have they ramped up clock-speed to ridiculous levels again to compete? Is that integer performance or FP? Is it the fact that they have increased to 8c/16 threads? Most mainstream apps have no hope of using these 16 threads as we know. Can't wait for some real world benchmarks. I will pray that this works out, we need some competition.

It's 40% clock for clock and that's 40% over AMD's new Excavator microarchitecture. If you compare it against something like Bulldozer or Piledriver CPU's (such as the FX 8350) Zen will be about 55 to 60% faster. If you look at some CPU graphs clock for clock Zen in theory will be on parity with Intel's Skylake and hopefully at a cheaper price.

in theory yes. Need some benchmarks, manufacturers always throw optimistic results based on best case scenario.

IPC is per clock. The statement X has a 40% IPC increase over Y means X at 3.0GHz is similar to Y at 4.2GHz. Matching clock for clock performance is great and all, but Intel is running in the 3-4GHz range so if Zen can't run that fast then the whole thing doesn't matter - they'd still be second fiddle to Intel's architecture.

yes, but more ipc vs a piledriver "module"? or one integer core? FP performance, something amd struggled with? I would like to be optimistic about this, but i was optimistic about bulldozer and that flopped, at least from a gaming perspective.

 

[TerryLaze]

yes, but more ipc vs a piledriver "module"? or one integer core? FP performance, something amd struggled with? I would like to be optimistic about this, but i was optimistic about bulldozer and that flopped, at least from a gaming perspective.

On their slide they claim that "the core" will have 40% more IPC ,but like the confusion you have about FX(up till now) it continues with zen since "the core" could be including the SMT,which I hear you say "of course it does" but if the core is so wide (in instructions) that outside of distributed computing you will never see the extra 40% being used without the SMT?

Their claims will be correct and the core will in deed have 40% higher IPC but the single thread speed,in anything a normal user would run,would still be abyssal.

 

[Onus]

So much speculation...

I built my AM3+ system based in part on the Bulldozer hype; I'm irritated anew every time I turn it on over how slow it is and/or how much power it sucks down. This weekend it may finally retire (assuming I get my small BTC wallet off of it).
I'd like to believe Zen will be fantastic, but I'm not going to buy into it unless or until actual benchmarks justify it.

 

[alextheblue]

The 5350 uses excavator architecture with jaguar cores, this is literally the closest thing to zen AMD has on the market currently. And they`re calling for a 40% improvement in IPC.

That's twice you've said this and it's wrong! Read my other post. Excavator is not Jaguar, Jaguar is not Excavator. They're not related. Different architectures, different families, etc etc.

 

[WizardlyGamer]

We are in desperate need of competition and innovation in the Desktop CPU front. Getting tired of mobile focused refreshes from Intel that give us little reason to upgrade.

We are in desperate need of competition and innovation in the Desktop CPU front. Getting tired of mobile focused refreshes from Intel that give us little reason to upgrade.

I just hate the fact that it takes 4 years for amd to upgrade their enthusiast line of processors, but its better than a yearly upgrade like intel, that gives 3% performance increase for $500 more.

 

[jimmysmitty]

We are in desperate need of competition and innovation in the Desktop CPU front. Getting tired of mobile focused refreshes from Intel that give us little reason to upgrade.

We are in desperate need of competition and innovation in the Desktop CPU front. Getting tired of mobile focused refreshes from Intel that give us little reason to upgrade.

I just hate the fact that it takes 4 years for amd to upgrade their enthusiast line of processors, but its better than a yearly upgrade like intel, that gives 3% performance increase for $500 more.

The only downside is that that performance boost AMD is getting is only going to be due to their poor performance right now.

And Yea 3% sucks. If only consumer software could use the extensions like AVX and properly use cores. Then we would see more than 3% each year from both sides.

 

[DonQuixoteMC]

<insert generic comment praising competition here>

but seriously though, it's great to see AMD back in the CPU game.

 

[InvalidError]

but seriously though, it's great to see AMD back in the CPU game.

They are only back on paper, we're still waiting to see what those claims will translate to in the real world. Until AMD's claims get independently confirmed, I'd stick to cautious optimism instead of gobbling AMD's hype whole.

 

[DonQuixoteMC]

They are only back on paper, we're still waiting to see what those claims will translate to in the real world. Until AMD's claims get independently confirmed, I'd stick to cautious optimism instead of gobbling AMD's hype whole.

True. They're definitely in the game, we'll see if they make it to the scoreboard.

 

[jimmysmitty]

They are only back on paper, we're still waiting to see what those claims will translate to in the real world. Until AMD's claims get independently confirmed, I'd stick to cautious optimism instead of gobbling AMD's hype whole.

True. They're definitely in the game, we'll see if they make it to the scoreboard.

The only rumor that has me worried for Zen right now is the clock speed. Because AMD is using Samsungs 14nm LPP node they might not be able to get high enough clocks. Then again it might also be another case of Athlon 3200+/6400+ etc and we might not have to worry.

It only worries me currently because I doubt they have something vastly faster per clock than Intel and Intel has a pretty decent 14nm.

 

[InvalidError]

It only worries me currently because I doubt they have something vastly faster per clock than Intel and Intel has a pretty decent 14nm.

I wouldn't worry about that. As long as AMD's performance bounces back to being a credible threat to mainstream i5 and i7, all AMD should need to do is provide better bang-per-buck to claw back some market share. If AMD could produce chips that perform similarly to Intel's and sold them for 20% less, AMD would still earn 20-30% more per chip than it does now.

 

[jimmysmitty]

It only worries me currently because I doubt they have something vastly faster per clock than Intel and Intel has a pretty decent 14nm.

I wouldn't worry about that. As long as AMD's performance bounces back to being a credible threat to mainstream i5 and i7, all AMD should need to do is provide better bang-per-buck to claw back some market share. If AMD could produce chips that perform similarly to Intel's and sold them for 20% less, AMD would still earn 20-30% more per chip than it does now.

True but if they perform similar they wont charge less and to me they need a real winner in the server market before the consumer market.

Clock speed only worries me because Intel has an advantage, especially in the i5 and i7s in the mainstream market.

Few more months and we shall see.

 

[InvalidError]

True but if they perform similar they wont charge less and to me they need a real winner in the server market before the consumer market.

When given the choice between the under-dog brand and the leading brand from the past several years for the same price, most people will choose the leading brand and completely ignore the under-dog. If AMD wants to regain market share, they will need to under-cut Intel. At price and performance parity, most of AMD's sales would go to AMD loyalists and not expand AMD's market share by much. Same goes for server chips: given the choice between AMD servers and Intel servers that cost and perform about the same, most companies would pick Intel.

When you lack the benefit of leading reputation, trying to charge the same as leading brands for a nearly identical product is often suicidal. It is much of the same between AMD and Nvidia: ~75% of gamers are going with Nvidia despite AMD GPUs with similar performance costing $50-100 less.